Open backnotprop opened 6 years ago
I've reverted to Cuda 8 and I get the same issue
This looks like a bug with PyTorch. I will convert the vgg weights later to avoid this issue.
@zhanghang1989 yea I figured, I submitted an issue to PyTorch as well.
Is there an alternative I can use for now?
I believe my download must've been corrupted ... Ive downloaded a separate vgg file and it seems to be working now... waiting to see how training goes
Great. please keep me posted.
@backnotprop what directory structure does the training expect? And are labels generated from the directory names or individual images?